Google Delays Gemini 3.5 Pro to July 17 for Ground-Up Architectural Rebuild

Google DeepMind has pushed back the launch of Gemini 3.5 Pro to July 17, opting to abandon the architecture that underpinned its previous Pro-tier models in favor of a ground-up redesign — a rare public slip in a release cadence Google has fought hard to keep aggressive.

Why the delay

According to reports, the rebuild targets three areas where internal testing showed the model lagging its rivals: mathematical reasoning, SVG scene generation and image quality. Google had cleared the model for a July launch as recently as late June, with a 2 million-token context window and an upgraded Deep Think mode among the headline features.

The competitive pressure is obvious. OpenAI's GPT-5.6 lineup — Sol, Terra and Luna — has reset expectations on agentic coding and long-horizon reasoning, while Anthropic's Fable 5 returned to global availability this month after its export-control pause and immediately reclaimed benchmark leads.

The stakes for Google

Gemini 3.5 Pro is the anchor for Google's enterprise AI push, feeding Workspace, Cloud and the consumer Gemini app that now counts more than 750 million users. A delayed but stronger model is a defensible trade; a delayed and merely competitive one would sharpen questions about whether Google's scale advantage in compute and data still translates into frontier leadership.

The company has not commented publicly on the revised date. If July 17 holds, the model will arrive into the most crowded frontier lineup the industry has ever seen — with GPT-5.6, Fable 5, Grok 4.5 and China's GLM-5.2 all shipping within weeks of one another.